Over time, music taste and consumption have become much more diverse and up to the user. Through on-demand streaming platforms such as Spotify and Apple Music, we are no longer restricted to radio or record labels. Independent artists have been enabled to publish and create by themselves, making an entrance into the industry much easier. With that, we have also seen a great amount of change in the modern listener. We now put much greater importance on listening habits and styles. Music has become a personal identifier, and we can track and categorize ourselves by what we consume. This is evident just from the pure popularity of Spotify Wrapped and how we share it across social media platforms.
In a paper by Beatrice Rammstedt and Oliver P. John, entitled “Measuring personality in one minute or less: A 10-item short version of the Big Five Inventory in English and German” Rammstedt and John work to better understand a user’s listening habits through a personality inventory test. This test was then scaled to apply to specific factors of the individual. The set of questions was then focused on the "Big 5" personality traits: Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. Then, translating these personality identifiers into listening habits, we can create connections between listening habits and personality traits.
We will first explore a general overview of how all of these traits affect factors including the mood and era of a playlist. Then, we will take a more granular look at each of the traits and how they relate to specific playlist features in each of the following tabs of this website. Notice that we individually explore Openness, Conscientiousness, and Extraversion; however, Agreeableness and Neuroticism are grouped toegther. This is because Agreeableness and Neuroticism are both more internally-oriented traits that dictate why people engage with music rather than how they engage with it.
We present our conclusions on the data at the end of this page, but we encourage you to explore the other website tabs first so you can form your own analysis of the data.
The data is broken into 3 sections: Demographics, Habits, and Personality. Demographics look at who the person is and their music they listen to. Habits correspond to lifestyle, athletic smoking etc. Lastly, Personality is the Big 5 traits relating to the paper by Rammstedt and John.
The dataset has over 10,000 playlists corresponding to 700+ different users. Therefore, personality has a one-to-many relationship with demographics, as multiple playlists correspond to one person with the same personality. There are over 130 attributes, but we focused on exploring 15 of them.
Five of the attributes we explored related to the user's Big 5 personality traits. Five of the other attributes related to the playlist's average valence, average energy, average song release year, average danceability, and the diversity of the artists in the playlist. The remaining five attributes related to the genres in each playlist. The dataset contained information on over 40 genres, so we filtered for only the top 5 genres (alternative, indie, local, pop, and rock) to explore the genre composition of the playlists. We used Pandas to identify understandable relationships that we visualized using Altair, D3, and Matplotlib.
Interact with the graph! Select a region of the scatterplot to see how each quadrant of music appeals to people with different dominant personality traits.
The scatterplot on the left represents the mood of the playlist. Valence refers to how positive or negative a playlist is, and energy refers to the energy level of the playlist. Therefore, the combination of valence and energy together create four mood quadrants on the scatterplot.
Looking at the music first at a large scale, we are able to draw some clear correlations. First, the scatter plot suggests a somewhat linear correlation between valence and energy; however, once we reach about 0.5 on valence and 0.7 in energy, the correlation appears to be more quadratic. Meaning that just because music has high energy, it does not necessarily make it positive. This directly relates to music such as hard rock or punk. Those genres of music are some of the loudest intense genres, but not directly “positive”.
Pivoting to our bar chart, we are also able to see on a large scale that the largest dominant trait of the participants is openness. It is important to note that the bar chart on the right does not include people who have multi-dominant personalities to highlight mood differences between people with only one dominant personality trait.
This bar chart has a clear correlation with the point made earlier regarding on-demand music. Listeners are empowered to explore different genres and styles of music in a way we have never seen before. It could be extremely interesting to see how these traits change with time. Does this spike in openness directly relate to music becoming more democratized?Now, looking at the time trended with music, we see this impact of more and more music being made in a faster and freer way. All traits, no matter what they are, experience a spike around the mid-2000s do to this surge in music; however, it becomes much more intricate when looking at the spikes in data. Most traits follow a similar pattern of a spike around the late 2010s; however, two do not follow this pattern: Conscientiousness and Openness. Both experience a valley between two peaks. Openness is a much larger gap, approximately 4-6 years. We can create certain connections between the political state of the country between 2012-2018, as well as how openness increased again, experiencing the largest spike out of any traits at around 2018-2019.
Overall, many of the conclusions that we came to surprised us and were in conflict with our hypothesis. It was extremely interesting to see how many traits we paired together with listening habits we predicted to have a positive correlation actually had negative relationships. The only outlier to that is Extraversion and danceability, which, even still, has a somewhat weak positive relationship.
It is also important to note how the saturation of genres impacts the music a user listens to. A playlist may be majority “pop” music, not because the user loves pop, but because of how broad a genre it is. The same idea also goes with the second most popular genre, rock, is it metal? blues? Or classic? Any rock listener could tell you how different those are from each other.
The final point of note is how all 5 personality traits can influence music preference. Isolating the Big 5 by each personality trait may lead to surprising results because a person is made up of the interactions between all five of these traits.
We quickly prototyped a random forest classifier to see if we could predict a person’s favorite genre out of the top 5 most popular genres based on their Big 5 traits. Although the model accuracy was relatively low (hovering around 50%), we chose this model for its interpretability in feature weights. As you can see, every Big 5 trait is approximately equal in importance.
This indicates that the combination of personality traits may be more telling of a person’s music taste than their individual traits (which is what we investigated). Furthermore, it shows how it’s hard to reduce a person to their traits in a way that can easily reveal their preferences. This truly does show how vast and interconnected we all are with the music we consume.